Logical Agents: Propositional Logic. Chapter 7

Transcription

1 Logical Agents: Propositional Logic Chapter 7

2 Outline Topics: Knowledge-based agents Example domain: The Wumpus World Logic in general models and entailment Propositional (Boolean) logic Equivalence, validity, satisfiability Inference rules and theorem proving forward chaining backward chaining resolution

3 Knowledge bases Inference engine Knowledge base domain independent algorithms domain specific content Knowledge base = set of sentences in a formal language Declarative approach to building an agent (or other system). Declarative: Sentences express assertions about the domain Knowledge base operations: Tell it what it needs to know Ask (itself?) what to do query Answers should follow from the contents of the KB

4 Knowledge bases Agents can be viewed: at the knowledge level i.e., what they know, regardless of how implemented at the implementation level (also called the symbol level) i.e., data structures and algorithms that manipulate them Compare: abstract data type vs. data structure used to implement an ADT.

5 A simple knowledge-based agent Function KB-Agent(percept) returns an action static: KB, a knowledge base t, a counter, initially 0, indicating time Tell(KB, Make-Percept-Sentence(percept, t)) action Ask(KB, Make-Action-Query(t)) Tell(KB, Make-Action-Sentence(action, t)) t t + 1 return action

6 A simple knowledge-based agent In the most general case, the agent must be able to: Represent states, actions, etc. Incorporate new percepts Update internal representations of the world Deduce hidden/implicit properties of the world Deduce appropriate actions

7 The Wumpus World 4 Stench Breeze 3 Breeze Stench Breeze Gold 2 Stench Breeze 1 Breeze Breeze START

8 Wumpus World PEAS description Performance measure: gold: +1000; death: -1000; -1 per step; -10 for using the arrow Environment: Squares adjacent to wumpus are smelly Squares adjacent to pit are breezy Glitter iff gold is in the same square Shooting kills wumpus if you are facing it Shooting uses up the only arrow Grabbing picks up gold if in same square Releasing drops the gold in same square Actuators: Left turn, Right turn, Forward, Grab, Release, Shoot Sensors: Breeze, Glitter, Smell, Bump, Scream

9 Observable:?? Wumpus world characterisation

10 Wumpus world characterisation Observable: No only local perception Deterministic:??

11 Wumpus world characterisation Observable: No only local perception Deterministic: Yes outcomes exactly specified Episodic:??

12 Wumpus world characterisation Observable: No only local perception Deterministic: Yes outcomes exactly specified Episodic: No sequential at the level of actions Static:??

13 Wumpus world characterisation Observable: No only local perception Deterministic: Yes outcomes exactly specified Episodic: No sequential at the level of actions Static: Yes Wumpus and pits do not move Discrete:??

14 Wumpus world characterisation Observable: No only local perception Deterministic: Yes outcomes exactly specified Episodic: No sequential at the level of actions Static: Yes Wumpus and pits do not move Discrete: Yes Single-agent:??

15 Wumpus world characterisation Observable: No only local perception Deterministic: Yes outcomes exactly specified Episodic: No sequential at the level of actions Static: Yes Wumpus and pits do not move Discrete: Yes Single-agent: Yes Wumpus is essentially a natural feature

16 Exploring a wumpus world OK OK A OK Percept: [Stench: No, Breeze: No, Glitter: No, Bump: No, Scream: No]

17 Exploring a wumpus world B A OK A OK OK Percept: [Stench: No, Breeze: Yes, Glitter: No, Bump: No, Scream: No]

18 Exploring a wumpus world P? B A OK P? A OK OK Percept: [Stench: No, Breeze: Yes, Glitter: No, Bump: No, Scream: No]

19 Exploring a wumpus world P? B A OK P? A OK S A OK Percept: [Stench: Yes, Breeze: No, Glitter: No, Bump: No, Scream: No]

20 Exploring a wumpus world P P? B OK A P? OK A OK S A OK W Percept: [Stench: Yes, Breeze: No, Glitter: No, Bump: No, Scream: No]

21 Exploring a wumpus world P P? B OK A P? OK A A OK S A OK W Percept: [Stench: No, Breeze: No, Glitter: No, Bump: No, Scream: No]

22 Exploring a wumpus world P P? OK B OK A P? OK A OK A OK S A OK W Percept: [Stench: No, Breeze: No, Glitter: No, Bump: No, Scream: No]

23 Exploring a wumpus world P P? OK B OK A P? OK A BGS OK A A OK S A OK W Percept: [Stench: Yes, Breeze: Yes, Glitter: Yes, Bump: No, Scream: No]

24 Tight spots P? B A OK P? P? A OK B A OK P? Breeze in (1,2) and (2,1) no safe actions

25 Tight spots P? B A OK P? P? A OK B A OK P? Breeze in (1,2) and (2,1) no safe actions If pits are uniformly distributed, (2,2) is more likely to have a pit than (1,3) + (3,1)

26 Tight spots S A Smell in (1,1) cannot safely move

27 Tight spots S A Smell in (1,1) cannot safely move Can use a strategy of coercion: shoot straight ahead wumpus was there dead safe wumpus wasn t there safe

28 Logic in the Wumpus World As the agent moves and carries out sensing actions, it performs logical reasoning. E.g.: If (1,3) or (2,2) contains a pit and (2,2) doesn t contain a pit then (1,3) must contain a pit. We ll use logic to represent information about the wumpus world, and to reason about this world.

29 Logic in general A logic is a formal language for representing information such that conclusions can be drawn The syntax defines the sentences in the language The semantics define the meaning of sentences; i.e., define truth of a sentence in a world E.g., in the language of arithmetic x + 2 y is a sentence; x2 + y > is not a sentence x + 2 y is true iff the number x + 2 is not less than y x + 2 y is true in a world where x = 7, y = 1 x + 2 y is false in a world where x = 0, y = 6

30 Semantics: Entailment Entailment means that one thing follows from another: KB = α Knowledge base KB entails sentence α if and only if: α is true in all worlds where KB is true Or: if KB is true then α must be true. E.g., the KB containing the Canucks won entails either the Canucks won or the Leafs won E.g., x + y = 4 entails 4 = x + y Entailment is a relationship between sentences (i.e., syntax) that is based on semantics Note: Brains (arguably) process syntax (of some sort).

31 Semantics: Models Logicians typically think in terms of models, which are complete descriptions of a world, with respect to which truth can be evaluated We say m is a model of a sentence α if α is true in m M(α) is the set of all models of α Thus KB = α if and only if M(KB) M(α) E.g. KB = Canucks won and Leafs won α = Canucks won M( ) x x x x x x x x x x x x x x x x x x x x x x x x x xx x xx x x x x x x x M(KB) x x x x x x x x x

32 Logic texts usually distinguish: Aside: Semantics an interpretation, which is some possible world or complete state of affairs, from a model, which is an interpretation that makes a specific sentence or set of sentences true. The text uses model in both senses (so don t be confused if you ve seen the terms interpretation/model from earlier courses). And if you haven t, ignore this slide! We ll use the text s terminology.

33 Entailment in the Wumpus World Consider the situation where the agent detects nothing in [1,1], moves right, detects a breeze in [2,1] Consider possible models for just the? s, assuming only pits? A B? A? With no information: 3 Boolean choices 8 possible models

34 Breeze Breeze Breeze Breeze Breeze Breeze Breeze Breeze Wumpus Models Consider possible arrangements of pits in [1,2], [2,2], and [3,1], along with observations:

35 Breeze Breeze Breeze Breeze Breeze Breeze Breeze Breeze Models of the KB: Wumpus Models KB KB = wumpus-world rules + observations

36 Breeze Breeze Breeze Breeze Breeze Breeze Breeze Breeze Wumpus Models KB KB = wumpus-world rules + observations α 1 = [1,2] is safe, KB = α 1, proved by model checking

37 Breeze Breeze Breeze Breeze Breeze Breeze Breeze Breeze Wumpus Models: Another Example KB KB = wumpus-world rules + observations

38 Breeze Breeze Breeze Breeze Breeze Breeze Breeze Breeze Wumpus Models: Another Example 2 2 KB KB = wumpus-world rules + observations α 2 = [2,2] is safe, KB = α 2

39 Inference In the case of propositional logic, we can use entailment to derive conclusions by enumerating models. This is the usual method of computing truth tables I.e. can use entailment to do inference. In first order logic we generally can t enumerate all models (since there may be infinitely many of them and they may have an infinite domain). An inference procedure is a (syntactic) procedure for deriving some formulas from others.

40 Inference Inference is a procedure for computing entailments. KB α = sentence α can be derived from KB by the inference procedure Entailment says what things are implicitly true in a KB. Inference is intended to compute things that are implicitly true.

41 Inference Inference is a procedure for computing entailments. KB α = sentence α can be derived from KB by the inference procedure Entailment says what things are implicitly true in a KB. Desiderata: Inference is intended to compute things that are implicitly true. Soundness: An inference procedure is sound if whenever KB α, it is also true that KB = α. Completeness: An inference procedure is complete if whenever KB = α, it is also true that KB α.

42 Propositional Logic: Syntax Propositional logic is a simple logic illustrates basic ideas We first specify the proposition symbols or (atomic) sentences: P 1, P 2 etc. Then we define the language: If S 1 and S 2 are sentences then: S 1 is a sentence (negation) S 1 S 2 is a sentence (conjunction) S 1 S 2 is a sentence (disjunction) S 1 S 2 is a sentence (implication) S 1 S 2 is a sentence (biconditional)

43 Propositional Logic: Semantics Each model assigns true or false to each proposition symbol E.g.: P 1,2 true, P 2,2 true, P 3,1 false (With these symbols, 8 possible models, can be enumerated.) Rules for evaluating truth with respect to a model m: S is true iff S is false S 1 S 2 is true iff S 1 is true and S 2 is true S 1 S 2 is true iff S 1 is true or S 2 is true S 1 S 2 is true iff S 1 is false or S 2 is true S 1 S 2 is true iff S 1 S 2 is true and S 2 S 1 is true Simple recursive process evaluates an arbitrary sentence, e.g., P 1,2 (P 2,2 P 3,1 ) = true (false true) = true true = true

44 Truth Tables for Connectives P Q P P Q P Q P Q P Q false false true false false true true false true true false true true false true false false false true false false true true false true true true true

45 Let P i,j be true if there is a pit in [i, j]. Wumpus World Sentences Let B i,j be true if there is a breeze in [i, j]. Information from sensors: P 1,1, B 1,1, B 2,1 Also know: pits cause breezes in adjacent squares

46 Let P i,j be true if there is a pit in [i, j]. Wumpus World Sentences Let B i,j be true if there is a breeze in [i, j]. Information from sensors: P 1,1, B 1,1, B 2,1 A square is breezy if and only if there is an adjacent pit B 1,1 (P 1,2 P 2,1 ) B 2,1 (P 1,1 P 2,2 P 3,1 ) Note: B 1,1 has no internal structure think of it as a string. So must write 1 formula for each square.

47 Let P i,j be true if there is a pit in [i, j]. Wumpus World Sentences Let B i,j be true if there is a breeze in [i, j]. Information from sensors: P 1,1, B 1,1, B 2,1 A square is breezy if and only if there is an adjacent pit B 1,1 (P 1,2 P 2,1 ) B 2,1 (P 1,1 P 2,2 P 3,1 ) Note: B 1,1 has no internal structure think of it as a string. So must write 1 formula for each square. Using logic can conclude P 1,2 and P 2.1 from B 1,1. Note, if you wrote the above as: B 1,1 (P 1,2 P 2,1 ) (I.e. A breeze implies a pit in an adjacent square ) you could not derive P 1,2 and P 2.1 from B 1,1. Crucial to express all information

48 Wumpus World KB For the part of the Wumpus world we re looking at, let KB = {R 1, R 2, R 3, R 4, R 5 } where R 1 is P 1,1 R 2 is B 1,1 (P 1,2 P 2,1 ) R 3 is B 2,1 (P 1,1 P 2,2 P 3,1 ) R 4 is B 1,1 R 5 is B 2,1

49 Truth Tables for Inference B 1,1 B 2,1 P 1,1 P 1,2 P 2,1 P 2,2 P 3,1 R 1 R 2 R 3 R 4 R 5 KB f f f f f f f t t t t f f f f f f f f t t t f t f f f t f f f f f t t f t t f f t f f f f t t t t t t t f t f f f t f t t t t t t f t f f f t t t t t t t t f t f f t f f t f f t t f t t t t t t t f t t f t f Enumerate rows (different assignments to symbols), For KB = α, if KB is true in row, check that α is too

50 Inference by Enumeration Function TT-Entails?(KB, α) returns true or false inputs: KB, the knowledge base, a sentence in propositional logic α the query, a sentence in propositional logic symbols a list of the proposition symbols in KB and α return TT-Check-All(KB, α, symbols, [ ])

51 Inference by Enumeration Function TT-Check-All(KB, α, symbols, model) returns true or false if Empty?(symbols) then if PL-True?(KB, model) then return PL-True?(α, model) else return true else do P First(symbols); rest Rest(symbols) return TT-Check-All(KB, α, rest, Extend(P, true, model)) and TT-Check-All(KB, α, rest, Extend(P, false, model)) Depth-first enumeration of all models Hence, sound and complete Algorithm is O(2 n ) for n symbols; problem is co-np-complete

52 Logical Equivalence Two sentences are logically equivalent iff true in same models: α β if and only if α = β and β = α The following should be familiar: (α β) (β α) (α β) (β α) ((α β) γ) (α (β γ)) ((α β) γ) (α (β γ)) ( α) α (α β) ( β α) (α β) ( α β) (α β) ((α β) (β α)) (α β) ( α β) (α β) ( α β) (α (β γ)) ((α β) (α γ)) (α (β γ)) ((α β) (α γ))

53 Validity and Satisfiability A sentence is valid if it is true in all models, e.g., A A, A A, (A (A B)) B

54 Validity and Satisfiability A sentence is valid if it is true in all models, e.g., A A, A A, (A (A B)) B Validity is connected to inference via the Deduction Theorem: KB = α if and only if (KB α) is valid

55 Validity and Satisfiability A sentence is valid if it is true in all models, e.g., A A, A A, (A (A B)) B Validity is connected to inference via the Deduction Theorem: KB = α if and only if (KB α) is valid A sentence is satisfiable if it is true in some model e.g., A B, C

56 Validity and Satisfiability A sentence is valid if it is true in all models, e.g., A A, A A, (A (A B)) B Validity is connected to inference via the Deduction Theorem: KB = α if and only if (KB α) is valid A sentence is satisfiable if it is true in some model e.g., A B, C A sentence is unsatisfiable if it is true in no models e.g., A A

57 Validity and Satisfiability A sentence is valid if it is true in all models, e.g., A A, A A, (A (A B)) B Validity is connected to inference via the Deduction Theorem: KB = α if and only if (KB α) is valid A sentence is satisfiable if it is true in some model e.g., A B, C A sentence is unsatisfiable if it is true in no models e.g., A A Satisfiability is connected to inference via the following: KB = α if and only if (KB α) is unsatisfiable I.e., prove α by reductio ad absurdum

58 Validity and Satisfiability A sentence is valid if it is true in all models, e.g., A A, A A, (A (A B)) B Validity is connected to inference via the Deduction Theorem: KB = α if and only if (KB α) is valid A sentence is satisfiable if it is true in some model e.g., A B, C A sentence is unsatisfiable if it is true in no models e.g., A A Satisfiability is connected to inference via the following: KB = α if and only if (KB α) is unsatisfiable I.e., prove α by reductio ad absurdum What often proves better for determining KB = α is to show that KB α is unsatisfiable.

59 Proof methods divide into (roughly) two kinds: 1. Application of inference rules: Proof Methods Legitimate (sound) generation of new sentences from old Proof = a sequence of inference rule applications. Can use inference rules as operators in a standard search algorithm. Typically require translation of sentences into a normal form 2. Model checking: Possibilities: Truth table enumeration (always exponential in n) Improved backtracking, e.g., DPLL Heuristic search in model space (sound but incomplete) e.g., min-conflicts hill-climbing algorithms

60 Specialised Inference: Forward and Backward Chaining We consider a very useful, restricted case: Horn Form KB = conjunction of Horn clauses Horn clause = proposition symbol; or (conjunction of symbols) symbol E.g., C, (B A), (C D B) Not: ( B A), (B A)

61 Horn clauses Technically a Horn clause is a clause or disjunction of literals, with at most one positive literal. I.e. of form A 0 A 1 A n or A 1 A n These can be written: A 1 A n A 0 or A 1 A n We won t bother with rules of the form A 1 A n Rules of this form are called integrity constraints. They don t allow new facts to be derived, but rather rule out certain combinations of facts.

62 Reasoning with Horn clauses Modus Ponens (for Horn form): Complete for Horn KBs α 1,..., α n, α 1 α n β β Can be used with forward chaining or backward chaining. These algorithms are very natural; forward chaining runs in linear time

63 KB: P Q, L M P, B L M, A P L, A B L, A, B Example

64 Forward chaining Idea: Fire any rule whose premises are satisfied in the KB, Add its conclusion to the KB, until query is found

65 Forward chaining algorithm Procedure: C := {}; repeat choose r A such that r is b 1 b m h b i C for all i, and h C; C := C {h} until no more choices

66 Forward chaining algorithm (from text) Function PL-FC-Entails?(KB,q) returns true or false inputs: KB the knowledge base, a set of propositional Horn clauses q the query, a proposition symbol local variables: count a table, indexed by clause, initially # of premises inferred a table, indexed by symbol, each entry initially false agenda a list of symbols, initially symbols known true in KB while agenda is not empty do p Pop(agenda) unless inferred[p] do inferred[p] true for each Horn clause c in whose premise p appears do decrement count[c] if count[c] = 0 then do if Head[c] = q then return true; Push(Head[c], agenda) return false

67 KB: P Q, L M P, B L M, A P L, A B L, A, B Query Q: Forward chaining example

68 KB: P Q, L M P, B L M, A P L, A B L, A, B Query Q: From A and B, conclude L Forward chaining example

69 KB: P Q, L M P, B L M, A P L, A B L, A, B Query Q: From A and B, conclude L From L and B, conclude M Forward chaining example

70 KB: P Q, L M P, B L M, A P L, A B L, A, B Query Q: From A and B, conclude L From L and B, conclude M From L and M, conclude P Forward chaining example

71 KB: P Q, L M P, B L M, A P L, A B L, A, B Query Q: From A and B, conclude L From L and B, conclude M From L and M, conclude P From P conclude Q Forward chaining example

72 Completeness FC, when run to completion, derives every atomic sentence entailed by KB 1. FC reaches a fixed point where no new atomic sentences are derived 2. Can consider the final state as a model m, assigning true/false to symbols 3. Claim: Every clause in the original KB is true in m Proof: Suppose a clause a 1... a k b is false in m Then a 1... a k is true in m and b is false in m Therefore the algorithm has not reached a fixed point! 4. Hence m is a model of KB 5. If KB = q, q is true in every model of KB, including m

73 Backward chaining We won t develop an algorithm for backward chaining here, but will just consider it informally. Idea with backward chaining: Start from query q and work backwards. To prove q by BC: check if q is known already; otherwise prove (by BC) all premises of some rule concluding q Avoid loops: Check if new subgoal is already on the goal stack Avoid repeated work: Check if new subgoal 1 has already been proved true, or 2 has already failed

74 KB: Backward chaining example P Q, L M P, B L M, A P L, A B L, A, B Query Q:

75 KB: Backward chaining example P Q, L M P, B L M, A P L, A B L, A, B Query Q: Establish P as a subgoal.

76 KB: Backward chaining example P Q, L M P, B L M, A P L, A B L, A, B Query Q: Establish P as a subgoal. Can prove P by proving L and M

77 KB: Backward chaining example P Q, L M P, B L M, A P L, A B L, A, B Query Q: Establish P as a subgoal. Can prove P by proving L and M For M: Can prove M if we can prove B and L

78 KB: Backward chaining example P Q, L M P, B L M, A P L, A B L, A, B Query Q: Establish P as a subgoal. Can prove P by proving L and M For M: Can prove M if we can prove B and L B is known to be true

79 KB: Backward chaining example P Q, L M P, B L M, A P L, A B L, A, B Query Q: Establish P as a subgoal. Can prove P by proving L and M For M: Can prove M if we can prove B and L B is known to be true L can be proven by proving A and B.

80 KB: Backward chaining example P Q, L M P, B L M, A P L, A B L, A, B Query Q: Establish P as a subgoal. Can prove P by proving L and M For M: Can prove M if we can prove B and L B is known to be true L can be proven by proving A and B. A and B are known to be true

81 KB: Backward chaining example P Q, L M P, B L M, A P L, A B L, A, B Query Q: Establish P as a subgoal. Can prove P by proving L and M For M: Can prove M if we can prove B and L B is known to be true L can be proven by proving A and B. A and B are known to be true For L: L can be proven by proving A and B. A and B are known to be true

82 KB: Backward chaining example P Q, L M P, B L M, A P L, A B L, A, B Query Q: Establish P as a subgoal. Can prove P by proving L and M For M: Can prove M if we can prove B and L B is known to be true L can be proven by proving A and B. A and B are known to be true For L: L can be proven by proving A and B. A and B are known to be true L and M are true, thus P is true, thus Q is true

83 Forward vs. backward chaining FC is data-driven, cf. automatic, unconscious processing, E.g., object recognition, routine decisions May do lots of work that is irrelevant to the goal Good for reactive agents

84 Forward vs. backward chaining FC is data-driven, cf. automatic, unconscious processing, E.g., object recognition, routine decisions May do lots of work that is irrelevant to the goal Good for reactive agents BC is goal-driven, appropriate for problem-solving, E.g., Where are my keys? How do I get a job? Complexity of BC can be much less than linear in size of KB Can also sometimes be exponential in size of KB Good for question-answering and explanation

85 General Propositional Inference: Resolution Resolution is a rule of inference defined for Conjunctive Normal Form (CNF) CNF: conjunction of disjunctions of literals A clause is a disjunctions of literals. E.g., (A B) (B C D). Write as: (A B), (B C D)

86 Resolution inference rule: l 1 l k, m 1 m n Resolution l 1 l i 1 l i+1 l k m 1 m j 1 m j+1 m n where l i and m j are complementary literals. (I.e. l i m j.) E.g., P 1,3 P 2,2, P 2,2 P 1,3 Resolution is sound and complete for propositional logic

87 Using resolution to compute entailments To show whether KB = α, show instead that KB α is unsatisfiable: 1 Convert KB α into conjunctive normal form. 2 Use resolution to determine whether KB α is unsatisfiable. 3 If so then KB = α; otherwise KB = α.

88 E.g.: B 1,1 (P 1,2 P 2,1 ) Conversion to CNF

89 E.g.: B 1,1 (P 1,2 P 2,1 ) Conversion to CNF 1 Eliminate, replacing α β with (α β) (β α). (B 1,1 (P 1,2 P 2,1 )) ((P 1,2 P 2,1 ) B 1,1 )

90 Conversion to CNF E.g.: B 1,1 (P 1,2 P 2,1 ) 1 Eliminate, replacing α β with (α β) (β α). (B 1,1 (P 1,2 P 2,1 )) ((P 1,2 P 2,1 ) B 1,1 ) 2 Eliminate, replacing α β with α β. ( B 1,1 P 1,2 P 2,1 ) ( (P 1,2 P 2,1 ) B 1,1 )

91 Conversion to CNF E.g.: B 1,1 (P 1,2 P 2,1 ) 1 Eliminate, replacing α β with (α β) (β α). (B 1,1 (P 1,2 P 2,1 )) ((P 1,2 P 2,1 ) B 1,1 ) 2 Eliminate, replacing α β with α β. ( B 1,1 P 1,2 P 2,1 ) ( (P 1,2 P 2,1 ) B 1,1 ) 3 Move inwards using de Morgan s rules and double-negation: ( B 1,1 P 1,2 P 2,1 ) (( P 1,2 P 2,1 ) B 1,1 )

92 Conversion to CNF E.g.: B 1,1 (P 1,2 P 2,1 ) 1 Eliminate, replacing α β with (α β) (β α). (B 1,1 (P 1,2 P 2,1 )) ((P 1,2 P 2,1 ) B 1,1 ) 2 Eliminate, replacing α β with α β. ( B 1,1 P 1,2 P 2,1 ) ( (P 1,2 P 2,1 ) B 1,1 ) 3 Move inwards using de Morgan s rules and double-negation: ( B 1,1 P 1,2 P 2,1 ) (( P 1,2 P 2,1 ) B 1,1 ) 4 Apply distributivity law ( over ) and flatten: ( B 1,1 P 1,2 P 2,1 ) ( P 1,2 B 1,1 ) ( P 2,1 B 1,1 )

93 Conversion to CNF E.g.: B 1,1 (P 1,2 P 2,1 ) 1 Eliminate, replacing α β with (α β) (β α). (B 1,1 (P 1,2 P 2,1 )) ((P 1,2 P 2,1 ) B 1,1 ) 2 Eliminate, replacing α β with α β. ( B 1,1 P 1,2 P 2,1 ) ( (P 1,2 P 2,1 ) B 1,1 ) 3 Move inwards using de Morgan s rules and double-negation: ( B 1,1 P 1,2 P 2,1 ) (( P 1,2 P 2,1 ) B 1,1 ) 4 Apply distributivity law ( over ) and flatten: ( B 1,1 P 1,2 P 2,1 ) ( P 1,2 B 1,1 ) ( P 2,1 B 1,1 ) For resolution, then write as ( B 1,1 P 1,2 P 2,1 ), ( P 1,2 B 1,1 ), ( P 2,1 B 1,1 )

94 Resolution Algorithm Function PL-Resolution(KB, α) returns true or false inputs: KB, the knowledge base, a sentence in propositional logic α, the query, a sentence in propositional logic clauses the set of clauses in CNF(KB α) new { } loop do if clauses contains the empty clause then return true if C i, C j are resolvable clauses where PL-Resolve(C i, C j ) clauses then clauses clauses PL-Resolve(C i, C j ) else return false Note that the algorithm in the text is buggy

95 Resolution Example E.g.: KB = (B 1,1 (P 1,2 P 2,1 )) B 1,1, α = P 1,2 Show KB = α by showing that KB α is unsatisfiable: P 2,1 B 1,1 B 1,1 P 1,2 P 2,1 P 1,2 B 1,1 B 1,1 P 1,2 B 1,1 P 1,2 B 1,1 P 1,2 P 2,1 P 1,2 B 1,1 P 2,1 B 1,1 P1,2 P 2,1 P 2,1 P 2,1 P 1,2

96 Show: Resolution: Another Example {r u, u w, r w} = w

97 Resolution: Continued There is a great deal that can be done to improve the basic algorithm: Unit resolution: propagate unit clauses (e.g. B 1,1 ) as much as possible. Note that this correspoinds to the minimum remaining values heuristic in constraint satisfaction! Eliminate tautologies Eliminate redundant clauses Eliminate clauses with literal l where the complement of l doesn t appear elsewhere. Set of support: Do resolutions on clauses with ancestor in α. Similar to backward chaining keep a focus on the goal.

98 Summary Logical agents apply inference to a knowledge base to derive new information and make decisions Basic concepts of logic: syntax: formal structure of sentences semantics: truth of sentences wrt models entailment: necessary truth of one sentence given another inference: deriving sentences from other sentences soundness: derivations produce only entailed sentences completeness: derivations can produce all entailed sentences

99 Summary (Continued) Wumpus world requires the ability to represent partial and negated information, reason by cases, etc. Forward, backward chaining are complete for Horn clauses. Forward chaining is linear-time for Horn clauses. Resolution is complete for propositional logic. Propositional logic lacks expressive power